First cost proof #

Status: experiment to validate; no complete public end-to-end proof is claimed here.

The owner-specified exact steady-state targets are $0.51/month for strall.com and $0.61/month for usekarma.dev, totaling $1.12/month. They describe the intended minimal state, not the cost of every historical Karma service. Their billing scope and assumptions require verification.

The objective is to explain every cent above that target and produce a safe remediation plan without destructive changes.

1. Derive expected instances and dependencies from versioned configuration and IaC; distinguish active intent from retained historical definitions.
2. Inventory relevant accounts and regions read-only. Record inaccessible scope, persistent resources, and discrepancies with Terraform/SSM metadata.
3. Retrieve billing evidence for an explicit period and cost basis. Separate billed spend from monthly forecasts; account for usage, shared charges, credits, taxes, and rounding as applicable.
4. Map charges to services/resources and back to configuration/IaC where possible. Preserve an unexplained residual when evidence is missing; never invent attribution.
5. Propose scoped remediation with expected monthly savings, dependencies, service impact, data-loss risk, recovery requirements, and required authorization.
6. Run deterministic checks and prepare relevant Terraform/Terragrunt plans. Missing access or failed plans remain blockers, not passing evidence.

The report should include repository revisions, observation timestamps, coverage, billing totals, target assumptions, an attribution ledger, resource ownership, proposed changes, check/plan results, and unresolved questions. Success requires reconciliation to billing precision with no unexplained excess and no unobserved scope silently treated as empty. If charges cannot be attributed or the target excludes unavoidable costs, report that result explicitly and revise assumptions with the owner.

No destructive operation is part of this proof. Savings remain estimates until a separately authorized change is followed by health checks and billing observation. Link a public, redacted evidence artifact here when one exists.

What comes next #

The future direction is repeatable desired/actual/cost reconciliation and broader component verification. Promotion depends on observed results, not agent confidence. The getting started remain the practical entry point; the agent workflow adds evidence and authority boundaries around infrastructure engineering.